9:31 p.m.
On 2023/05/12 0:32, Rongwei Wang wrote:
> btw, how can I create this situation seeing zero page easily?
> I also would like to test the patch.
Here a testcase that I used to allocate zero page:
Thanks! I could see it on x86_64.
crash> vtop -c 638970 7f018fb11000
VIRTUAL PHYSICAL
7f018fb11000 3af648000
PGD: 4646ca7f0 => 5b1ba7067
PUD: 5b1ba7030 => 5f9e63067
PMD: 5f9e633e8 => b45b7d067
PTE: b45b7d888 => 80000003af648225
PAGE: 3af648000 (ZERO PAGE)
z.c:
#include <string.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <ctype.h>
unsigned long *zero_data;
void read_only(void)
{
unsigned long i, val, count = 0;
printf("hello\n");
while (count++ < 10) {
for (i=0; i<(4 * 1024 * 1024) / sizeof(unsigned long); i+=1024) {
val = zero_data[i];
}
}
return;
}
int main(int argc, char *argv[])
{
char c;
zero_data = (unsigned long *)malloc(1024 * 1024 * 4);
if (!zero_data) {
printf("malloc fail: exit\n");
return -1;
}
printf("zero: %lx\n", zero_data);
read_only();
c = getchar();
free(zero_data);
exit(EXIT_SUCCESS);
}
When you test in arm64 with 64k base pagesize, the size of allocating memory is needed to
set more than 512MB.
I see.
>
>>> PTE PHYSICAL FLAGS
>>> 160000836e71fc3 836e71000
(VALID|USER|RDONLY|SHARED|AF|NG|PXN|UXN|SPECIAL)
>>>
>>> VMA START END FLAGS FILE
>>> ffff000844f51860 ffff8917c000 ffff8957d000 100073
>>>
>>> PAGE PHYSICAL MAPPING INDEX CNT FLAGS
>>> fffffe001fbb9c40 836e71000 0 0 1 2ffffc000001000
reserved
> a bit verbose, please replace these with "..."
Your mean comment as below is OK?
Yes.
crash> vtop -c 13674 ffff8917e000
VIRTUAL PHYSICAL
ffff8917e000 836e71000
PAGE DIRECTORY: ffff000802f8d000
PGD: ffff000802f8dff8 => 884e29003
PUD: ffff000844e29ff0 => 884e93003
PMD: ffff000844e93240 => 840413003
PTE: ffff000800413bf0 => 160000836e71fc3
PAGE: 836e71000 (ZERO PAGE)
...
>>> + if
(readmem(symbol_value("huge_zero_pfn"), KVADDR,
>>> + &huge_zero_pfn, sizeof(huge_zero_pfn),
>>> + "read huge_zero_pfn", QUIET|RETURN_ON_ERROR))
>>> + vt->huge_zero_paddr = huge_zero_pfn << PAGESHIFT();
> In kernel, huge_zero_pfn is initialized as ~0UL and set to a pfn when
> get_huge_zero_page() is called.
>
> crash> p -x huge_zero_pfn
> huge_zero_pfn = $4 = 0xffffffffffffffff
>
> so if huge_zero_pfn is ~0UL, vt->huge_zero_paddr should not be set.
Good catch.
How about following modification:
+ if (readmem(symbol_value("huge_zero_pfn"), KVADDR,
+ &huge_zero_pfn, sizeof(huge_zero_pfn),
+ "read huge_zero_pfn", QUIET|RETURN_ON_ERROR))
+ vt->huge_zero_paddr = (huge_zero_pfn != ~0UL) ? (huge_zero_pfn <<
PAGESHIFT());
hmm, this looks wrong, needs ':'?
I would prefer this simply:
if (readmem(symbol_value("huge_zero_pfn"), KVADDR,
&huge_zero_pfn, sizeof(huge_zero_pfn),
- "read huge_zero_pfn", QUIET|RETURN_ON_ERROR))
+ "read huge_zero_pfn", QUIET|RETURN_ON_ERROR)
&&
+ huge_zero_pfn != ~0UL)
vt->huge_zero_paddr = huge_zero_pfn << PAGESHIFT();
>
>>> + }
>>> +
>>> if (vt->flags & PERCPU_KMALLOC_V1)
>>> vt->dump_kmem_cache = dump_kmem_cache_percpu_v1;
>>> else if (vt->flags & PERCPU_KMALLOC_V2)
>>> diff --git a/x86_64.c b/x86_64.c
>>> index 5019c69..693a08b 100644
>>> --- a/x86_64.c
>>> +++ b/x86_64.c
>>> @@ -2114,8 +2114,9 @@ x86_64_uvtop_level4(struct task_context *tc, ulong
uvaddr, physaddr_t *paddr, in
>>> goto no_upage;
>>> if (pmd_pte & _PAGE_PSE) {
>>> if (verbose) {
>>> - fprintf(fp, " PAGE: %lx (2MB)\n\n",
>>> - PAGEBASE(pmd_pte) & PHYSICAL_PAGE_MASK);
>>> + fprintf(fp, " PAGE: %lx (2MB%s)\n\n",
>>> + PAGEBASE(pmd_pte) & PHYSICAL_PAGE_MASK,
>>> + IS_ZEROPAGE(PAGEBASE(pmd_pte) & PHYSICAL_PAGE_MASK) ?
", ZERO PAGE" : "");
>>> x86_64_translate_pte(pmd_pte, 0, 0);
>>> }
>>> @@ -2143,8 +2144,8 @@ x86_64_uvtop_level4(struct task_context *tc, ulong
uvaddr, physaddr_t *paddr, in
>>> *paddr = (PAGEBASE(pte) & PHYSICAL_PAGE_MASK) +
PAGEOFFSET(uvaddr);
>>> if (verbose) {
>>> - fprintf(fp, " PAGE: %lx\n\n",
>>> - PAGEBASE(*paddr) & PHYSICAL_PAGE_MASK);
>>> + fprintf(fp, " PAGE: %lx %s\n\n", PAGEBASE(*paddr) &
PHYSICAL_PAGE_MASK,
>>> + IS_ZEROPAGE(PAGEBASE(*paddr) & PHYSICAL_PAGE_MASK) ?
"(ZERO PAGE)" : "");
>>> x86_64_translate_pte(pte, 0, 0);
>>> }
> why only for uvtop_level4?
That because I have no idea about xyz_xen_wpt and similar target. In addition, I have no
environment to test that, so...
How about doing this when someone need it in these environment? Or I can do it if you
think just complete it directly is ok.
Ah ok, I see. Your patch is good enough for now.
Thanks,
Kazu